Infrastructure Sharding: Region-based Goval Projects

Replit's core challenge is ensuring that Repls (Linux containers) run reliably. To this end, it operates Goval infrastructure consisting of virtual machines, databases, cloud storage, and reverse proxies.

The existing infrastructure operated within a single failure domain, creating the risk that failures could spread across the entire system. In addition, the user membership-based partitioning method caused imbalance between clusters, and a control plane concentrated in a specific region was structurally vulnerable to regional outages.

To address this, Replit is redesigning its infrastructure from the ground up. The main strategies are as follows:

• Introduce Consistent Hashing to keep cluster sizes even and distribute load efficiently.
• Use IaC (Infrastructure as Code) to unify infrastructure management and configuration, enabling clusters to be created and removed on demand.
• Separate failure domains into smaller, more even units to strengthen the overall reliability of the system.

This structural change not only improves operational efficiency but also enables building dedicated Single-tenant clusters for enterprise customers.